[PATCH] PR 78534, 83704 Large character lengths

Janne Blomqvist blomqvist.janne@gmail.com
Sun Jan 14 10:46:00 GMT 2018


On Sat, Jan 13, 2018 at 7:35 PM, Dominique d'Humières
<dominiq@tournesol.lps.ens.fr> wrote:
> I have finally bootstrapped gfortran with the two patches applied and the spurious warnings with -Wall are now gone (limited testing), but I see a regression for gfortran.dg/string_1.f90 due to an additional error
>
> /opt/gcc/_clean/gcc/testsuite/gfortran.dg/string_1.f90:13:15:
>
>    print *, len(s)
>                1
> Error: Result of LEN overflows its kind at (1)
>
> No idea why this was not emitted without the patch.

Thanks for testing, I'll look into this. Was it on a 32 or 64-bit target?

(I suspect the issue could be that the 32-bit truncated value fits in
[0,INT_MAX] and hence no problem is detected, but I'll have to look
into it further to say something definite).

>> Le 12 janv. 2018 à 21:43, Joseph Myers <joseph@codesourcery.com> a écrit :
>>
>> On Fri, 12 Jan 2018, Janne Blomqvist wrote:
>>
>>> (I don't know why HOST_WIDE_INT_PRINT_DEC doesn't work with the
>>> diagnostics machinery on darwin, but IMHO at this point it's too late
>>
>> HOST_WIDE_INT_PRINT_DEC is a host-specific printf format; for example, it
>> might use %I64d on Windows host or %lld elsewhere.  The formats used in
>> diagnostic functions must *not* be host-specific or use macros anywhere in
>> them at all, because that won't work with translation.  Diagnostic
>> functions have their own host-independent formats; %wd is the correct
>> format for HOST_WIDE_INT in diagnostic functions.
>>
>> --
>> Joseph S. Myers
>> joseph@codesourcery.com
>
> I have tried to use %wp, but it didn’t work:
>
> ../../work/gcc/fortran/decl.c: In function 'void gfc_set_constant_character_len(gfc_charlen_t, gfc_expr*, gfc_charlen_t)':
> ../../work/gcc/fortran/decl.c:1567:5: error: unknown conversion type character 'w' in format [-Werror=format=]
>      "CHARACTER expression at %L is being truncated "
>      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>      "(%wd/%wd)", &expr->where,
>      ~~~~~~~~~~~
> ../../work/gcc/fortran/decl.c:1567:5: error: unknown conversion type character 'w' in format [-Werror=format=]
> ../../work/gcc/fortran/decl.c:1567:5: error: too many arguments for format [-Werror=format-extra-args]
> ../../work/gcc/fortran/decl.c:1575:17: error: unknown conversion type character 'w' in format [-Werror=format=]
>   gfc_error_now ("The CHARACTER elements of the array constructor "
>                  ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>           "at %L must have the same length (%wd/%wd)",
>           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> ../../work/gcc/fortran/decl.c:1575:17: error: unknown conversion type character 'w' in format [-Werror=format=]
> ../../work/gcc/fortran/decl.c:1575:17: error: too many arguments for format [-Werror=format-extra-args]

Yes, same for the on x86-64 Linux. Might be something weird on the
gfortran diagnostics stuff, as it seems %wd is used in the middle end.
Anyway, that's stage1 stuff IMHO.



-- 
Janne Blomqvist



More information about the Gcc-patches mailing list